AWS Certified Generative AI Developer – Professional (AIP-C01): the exam, explained

Short answer

The AWS Certified Generative AI Developer – Professional (exam code AIP-C01) is AWS's professional-tier certification for engineers who put foundation models into production — RAG and vector stores, agentic systems, safety and governance, cost and operations, and evaluation, on Bedrock and the services around it. It is 75 questions in 180 minutes, costs 300 USD, and is scored 100–1,000 with a 750 pass mark. Its five domains are weighted 31 / 26 / 20 / 12 / 11.

It is a Professional exam and it behaves like one: AWS pitches it at people with two or more years building on AWS plus a year of hands-on GenAI work. If that is not you yet, the foundational AI Practitioner (AIF-C01) is the on-ramp. Below: what it is, the domains, the format and cost, who should wait, what it actually tests versus what you already know, how it compares to AIF-C01 and Anthropic's CCAR-F, and how our own Track D maps onto it.

What is the AWS Certified Generative AI Developer – Professional?

AIP-C01 is AWS's professional-level credential for the GenAI developer role. The official exam guide frames it as validating that you can integrate foundation models into applications and business workflows and take GenAI solutions to production on AWS — the exam-guide list runs from vector stores, RAG and knowledge bases through prompt engineering, agentic AI, cost and performance optimisation, security and Responsible AI, to monitoring, troubleshooting and model evaluation. Model development and training, advanced ML techniques, and data or feature engineering are explicitly out of scope: this is a builder-and-operator exam, not a data-science one.

The name is the first trap. It is one letter away from the AI Practitioner (AIF-C01), and the two share nothing but the vendor: AIF-C01 is a foundational literacy exam with a 700 pass mark; AIP-C01 is the professional building exam with a 750 pass mark, its own five domains and twice the seat time. The side-by-side blueprints for the exams a senior engineer might weigh live in the AI certification exam guide; this page is the AIP-C01 walkthrough.

Exam domains and weights

Five content domains. Domain 1 — foundation-model integration, data management and compliance — is nearly a third of the scored content on its own, and Domains 1 and 2 together are 57% of it. The tail (Domains 4 and 5) is under a quarter combined.

The domain titles and weights are taken from the content outline of the AIP-C01 exam guide, which states each as a percentage of scored content. Because scoring is compensatory (below), a weak domain is survivable — but a weak Domain 1 is a lot harder to survive than a weak Domain 5.

Format, cost, pass mark

Everything in this table is stated on AWS's own pages, so it is confirmed rather than inferred; cost and seat time can still vary by region and change over time, so reconfirm at registration.

Two figures do the real work here. 75 questions in 180 minutes is 2.4 minutes per item, and Professional-tier AWS items are scenario paragraphs with several plausible services in the options — the time is adequate, not generous. And multiple response gives no partial credit: per the exam guide, you must select every correct response to score the item, so a four-fifths-right answer is a wrong answer. Cost, duration, delivery and languages come from the AWS certification page for the exam.

Who should not sit it yet

Straight answer: if you have not shipped a GenAI feature to production on AWS, do not book this exam. The exam page pitches it at someone who has shipped production software on AWS (or on open-source stacks) for a couple of years, has some AI/ML or data-engineering background, and has spent about a year actually building GenAI features — not studying them. That is a description of the person the questions were written for, not a formality — the scenarios assume you already know what a Bedrock Knowledge Base, provisioned throughput, or a Step Functions state machine is for, and test whether you pick the right one under a stated constraint.

• New to AWS? Start with the AWS fundamentals, not this exam. AWS itself lists the AI Practitioner (AIF-C01), Solutions Architect – Associate, Machine Learning Engineer – Associate and Data Engineer – Associate as certifications a candidate could benefit from earning first — none is required, but the list tells you what the item writers assume you know.
• Strong engineer, new to GenAI? The concepts (RAG, embeddings, agent loops, evals) are learnable in weeks; the AWS-specific service mapping and the exam's operating vocabulary are what take time. Build one real thing on Bedrock first. The AIF-C01 guide covers the on-ramp exam if you want a checkpoint on the way.
• Deep GenAI engineer, not on AWS? You will know the patterns and not the service names. A large share of the items come down to "which AWS service, configured how" — budget for the catalogue, not just the concepts.

The senior who should sit it: someone who has run GenAI in production on AWS and wants the vendor-recognised signal for that work — especially if you are aiming at AWS-partner GenAI roles, where this is the credential you would expect job descriptions to name. Our broader take on ROI, including when a cert is a distraction, is in which AI certifications are worth it.

What the exam actually tests vs what you already know

If you are already building GenAI systems, the exam's concept list will read as your day job: the guide's technologies-and-concepts page names RAG, vector databases and embeddings, prompt engineering and management, agentic AI, content safety, model evaluation, cost and performance for AI workloads, observability, and the surrounding platform concerns (event-driven architecture, serverless, containers, IaC, CI/CD). None of that is exotic to a working GenAI engineer. What the exam actually differentiates on is narrower:

• The AWS service map, at the "which one, and why not the others" level. The in-scope services list runs well past the obvious Bedrock family (Bedrock, AgentCore, Knowledge Bases, Prompt Management, Prompt Flows) into SageMaker AI and its sub-services, Kendra, OpenSearch, Step Functions, EventBridge, API Gateway, Lambda, ECS/EKS, KMS, IAM, CloudWatch, X-Ray, Cost Explorer and more. A distractor is usually a real service that solves an adjacent problem.
• Exam-named agent tooling. Domain 2's agentic task points at Strands Agents and Agent Squad when the work spans several agents, treats MCP as the way an agent reaches a tool, and leans on Step Functions for orchestration, stopping conditions and human review. If your agent stack is LangGraph or the Anthropic SDK, the concepts transfer and the names do not — learn the AWS-native equivalents by name.
• Operating constraints you may not have hit yet. Provisioned throughput versus on-demand, model cascading to cheaper models, prompt and response caching, per-token cost anomaly detection, guardrail placement, data-residency and compliance controls on the retrieval path. A senior who has only run a pilot has opinions here; the exam wants the AWS-documented behaviour.
• Reading the constraint in the stem. Professional-tier AWS items hinge on one qualifier — "lowest operational overhead", "most cost-effective", "must remain within the VPC" — and several options are technically correct without it. This is a reading discipline, and it is where experienced engineers lose marks by answering the question they would have asked.

Put differently: the durable skills are the ones you have; the exam-specific work is mapping them onto AWS's names and defaults, and rehearsing the reading discipline under a 2.4-minute clock.

AIP-C01 vs AIF-C01 vs CCAR-F — for someone choosing

Three credentials come up together for a senior engineer moving into AI. They are not alternatives so much as different questions.

• AIF-C01 or AIP-C01? A tiering question. If you build, and you have the AWS years, skip AIF-C01 and sit AIP-C01. AIF-C01 earns its place only if AI is genuinely new to you or you want a low-stakes first AWS AI exam.
• AIP-C01 or CCAR-F? A platform question, not a level question. AIP-C01 is about the AWS platform around any model; CCAR-F is about designing with Claude specifically and is scenario-based rather than service-catalogue-based. An engineer on AWS-partner work wants AIP-C01; one whose product is built on Claude wants CCAR-F; plenty of seniors end up wanting both, and the concept overlap (agentic loops, tool design, evals, safety) is real. The Claude exam has its own page: the Anthropic certification guide.

How to prepare

• Weight your time like the exam does. D1 (31%) and D2 (26%) are 57% of the marks: RAG and knowledge bases, vector stores, data pipelines, prompt management, agentic implementation, deployment and enterprise integration. Get those to reflex before polishing D4/D5.
• Learn the service catalogue by contrast. For every in-scope service, be able to say what its nearest neighbour is and when you would pick the other one (Knowledge Bases vs Kendra vs OpenSearch; Lambda vs ECS for an MCP server; provisioned throughput vs on-demand).
• Practise multiple response as all-or-nothing. Treat each option as its own true/false and only commit when every one is decided — there is no partial credit.
• Rehearse the clock. 2.4 minutes per item, and long stems. Read the last sentence of the scenario first; it usually holds the qualifier that eliminates half the options.
• Read the domain pages of the exam guide, not a summary of them. The task and skill statements name the services and patterns the item writers were working from.

If your real target is the work rather than the certificate, build against the domains directly and let the exam fall out as a checkpoint — which is how the AWS track of aiArch's curriculum is designed, and the next section shows the mapping.

As-built: how Track D maps to the AIP-C01 domains

This is not exam-prep advice; it is the actual coverage record from docs/CURRICULUM.md, which assigns each of Track D's 22 modules to one AIP-C01 domain and, underneath that, to the specific official task statements the module was written against. Module count per domain roughly tracks domain weight, which is a deliberate backward-design call: D1 at 31% gets six modules; D5 at 11% gets two.

Two things we learned building it that a generic prep list will not tell you. First, the exam is exam-weighted, not durable-skill-weighted: two of the modules — trackD.m8a (enterprise integration architectures) and trackD.m10a (governance and compliance) — are essentially AWS-service checklists, high exam yield and low durable-skill value, and the curriculum's own learning-path note schedules them for the final exam-prep week rather than the core path. If you are learning the job and not just sitting the exam, that is the honest order. Second, Domain 4's five modules over-represent its 12%, and the curriculum says so: trackD.m11a (intelligent caching) and trackD.m12a (quality and troubleshooting) are flagged as merge candidates into m11 and m12 for a future content pass, kept separate for now only because learner mastery records key on the existing lesson ids. Cost, caching, performance and observability are what a senior gets asked about in production long after the exam, so the depth is not wasted — but the module count there is a build artefact, not a signal about the exam. Track D also leads Domain 2 with the exam-named agent tooling (Strands, MCP, Step Functions) and uses Bedrock AgentCore as the production runtime for its applied work, because that is what the AWS-partner GenAI consultancies this track targets actually run.

Frequently asked questions

How hard is the AWS Certified Generative AI Developer – Professional exam?

Professional-tier, and pitched that way: a 750/1,000 scaled pass mark (the foundational AIF-C01 is 700), 75 scenario-style questions in 180 minutes, multiple-response items with no partial credit, and a target candidate AWS describes as having two-plus years building production applications on AWS and a year of hands-on GenAI work. The concepts are not the hard part for a working GenAI engineer; the AWS service mapping, the exam-named tooling (Strands, Agent Squad, MCP, Step Functions, AgentCore) and the read-the-qualifier discipline are. Scoring is compensatory, so one weak domain does not fail you on its own.

How many questions are on the AIP-C01 exam, and how long is it?

75 questions in 180 minutes, per AWS's exam page. The exam guide breaks the 75 into 65 scored items and 10 unscored pretest items that AWS uses to evaluate future questions; they are not identified, so answer everything. Unanswered questions score as incorrect and there is no penalty for guessing.

How much does the AIP-C01 exam cost?

300 USD, per AWS's exam page, delivered at a Pearson VUE test centre or online proctored, in English, Japanese, Korean or Simplified Chinese. AWS's exam-pricing page carries regional and foreign-exchange detail, and prices change, so confirm at registration.

AIF-C01 or AIP-C01 — which should I take?

If you build software on AWS and have the years, skip the AI Practitioner (AIF-C01) and go straight to AIP-C01; AWS itself says no prior certification is required. Take AIF-C01 first only if AI is genuinely new to you, you want a low-stakes first AWS AI exam, or your role is adjacent to the building rather than doing it. They are different tiers with different domains, not two versions of one exam. The AIF-C01 guide covers the on-ramp in the same detail as this page.

Is the AWS Generative AI Developer – Professional certification worth it?

For a senior engineer targeting AWS-partner or AWS-heavy GenAI work, yes: it is the vendor-recognised credential for exactly the role, and the domains (RAG, agentic implementation, safety and governance, cost and operations, evaluation) are the durable skills of the job, so preparing for it is not wasted even if the certificate never matters. For an engineer whose product is not on AWS, the platform-specific third of the exam is sunk cost; Anthropic's CCAR-F or simply shipping may be the better use of the same weeks. Our full ROI reasoning is in which AI certifications are worth it.

Do I need the AI Practitioner or another AWS cert before AIP-C01?

No. AWS states no specific certification is required beforehand, while noting candidates could benefit from the AI Practitioner, Solutions Architect – Associate, Machine Learning Engineer – Associate or Data Engineer – Associate first. Read that list as a description of the assumed background — core AWS services, IAM and security practice, deployment and IaC tooling, monitoring, cost principles — rather than a prerequisite ladder.

Sources & provenance

• AWS Certified Generative AI Developer – Professional (AIP-C01) exam guide — target candidate, out-of-scope tasks, question types (multiple choice / multiple response, all-correct rule, no penalty for guessing), 65 scored + 10 unscored, scaled 100–1,000, pass mark 750, compensatory scoring, and the five domain titles and weights (31 / 26 / 20 / 12 / 11). Fetched 2026-08-15.
• AWS certification page for the exam — Professional level, 180 minutes, 75 questions, 300 USD, Pearson VUE / online proctored, four languages, target-candidate experience, and the "could benefit from" prior-certification list. Fetched 2026-08-15; cost and seat time can vary by region and date.
• Exam guide sub-pages — technologies and concepts, in-scope AWS services, and the Domain 2 task statements (Strands Agents, AWS Agent Squad, MCP, Step Functions). Paraphrased on this page, not reproduced.
• Disambiguation: AIF-C01 (AWS Certified AI Practitioner, pass 700, 65 q / 90 min / 100 USD) and Anthropic CCAR-F (pass 720, 60 q / 120 min / 125 USD, public v1.0 exam guide) are separate exams — see the AIF-C01 guide, the CCAR-F guide and the exam guide.
• As-built: docs/CURRICULUM.md Track D section — the 22-module domain mapping and the learning-path note on exam-week-only modules.

Every exam figure on this page was checked against the two AWS pages above on 2026-08-15. No sample questions appear here by design — we do not reproduce or imitate exam content. Corrections: hello@aiarch.dev.

Originally published at aiarch.dev/aws-generative-ai-developer-certification, where it is kept up to date.

Want the skeleton instead of the essay? aiarch-templates has the src/lib/ seams, a threshold-gated eval stub and a cost-model skeleton. It is deliberately empty — it fixes the shape and you write the implementation. Apache-2.0.